A community-driven approach to enhancing the quality and interpretability of microscopy images

Since the advent of the first microscopes, biologists have harboured a profound desire to observe and document the living world. At present, around a million new publications are indexed in PubMed every year, of which ∼20% include figures with image data, such as microscopy, diagnostic and radiology images (Lee et al., 2018). Notably, cell biologists lead the way in terms of the sheer volume of image-based figures they produce. Given the ever expanding array of microscopy techniques available to life scientists (Hell, 2007; Huisken and Stainier, 2009; Wassie et al., 2019), this trend is expected to continue on its upward trajectory.

The FAIR standards (Wilkinson et al., 2016), which emphasize that scientific data is findable, accessible, interoperable and reproducible, have crystallized the prerequisites for research data, including image data, and have been embraced by the scientific community. Central to preserving the enduring value of images is the imperative that they are accessible to the community, for example, in papers and collections. To this end, the publication of images with comprehensive methodological explanations represents a crucial first step. Facilitating the reuse of images constitutes yet another pivotal stride forward. The recent development of image repositories and archives has played a key role in achieving this aim (Ellenberg et al., 2018; Swedlow et al., 2021). General repositories, such as Zenodo, OSF or OMERO servers, and specialized, fully searchable image databases, such as the BioImage Archive (Hartley et al., 2022), EMPIAR for electron microscopy datasets, or the Image Data Resource, with reference image datasets, are also emerging as important community resources.

However, despite this progress, many images in publications fall short of meeting the FAIR criteria. A significant number of publications lack essential details regarding image acquisition or processing methods, thereby impeding reproducibility (Marques et al., 2020). Furthermore, the scientific community is increasingly vigilant of the presence of intentionally or inadvertently misleading images in publications (Bik et al., 2016; Rossner and Yamada, 2004). Indeed, Bik et al. reported an alarming increase in the incidence of problematic images, with their survey revealing that ∼4% of published papers contain misleading images, half of which appear to be deliberate falsifications. To avoid misleading image figures, many scientists and communities have published valuable resources (Bik et al., 2016, 2018; Cromey, 2013; CSE, 2012; North, 2006; Rossner and Yamada, 2004) and workflows (Schmied and Jambor, 2020; Senft et al., 2023) for preparing truthful image figures. In the age of electronic image data, there is also a growing emphasis on appropriate image handling and analysis (Aaron and Chew, 2021; Hammer et al., 2021; Martin and Blatt, 2013; Miura and Norrelykke, 2021).

Figures often serve as the initial focal point for readers when they interact with a publication (see article by E. Pain at https://www.science.org/content/article/how-seriously-read-scientific-paper). Consequently, images within figures should not only avoid any form of misrepresentation but should also be presented in a manner that is entirely comprehensible. However, at present many images still lack interpretability. A recent community-led project, which assessed image quality in high impact publications spanning cell biology, plant sciences and physiology, identified a number of recurring issues that act as major impediments to understandable image figures (Jambor et al., 2021). These include missing scale bars, unclear insets, insufficient annotations of symbols and the use of colours that are inaccessible to colourblind readers. In summary, only 2% of plant science, 12% of cell biology and 16% of physiology papers fulfilled all good practice criteria (Jambor et al., 2021). The consequence of subpar figures is that data remains enigmatic or confusing, and the biological insights remain shrouded in obscurity to readers.

Here, I outline our recent community-driven initiative that aims to improve the quality of images and image analysis figures in publications. I also describe how this effort seamlessly aligns with existing initiatives in the realms of microscopy data. Finally, I glance back at historical discourse and community efforts that have revolved around scientific images.

“A picture is worth a thousand words”.

This phrase is only true if the picture in question is understandable to the audience, or else it might spark a thousand questions. Every training course in data visualization invariably discusses, for example, colour usage and the importance of accessible – including to colourblind – choices (Cairo, 2012; Nussbaumer Knaflic, 2015; Tufte, 2011). In fact, even in the 1939 textbook on graphic presentations there was a full chapter on ‘Color and its use’, which cautioned the audience to avoid combining red and green data encodings (Brinton, 1939). Colour perception and further fundamentals of information design of course also apply to image figures; see, for example, the ASCB guidelines (https://www.ascb.org/science-news/how-to-make-scientific-figures-accessible-to-readers-with-color-blindness/) or a recently published technology feature (Katsnelson, 2021). I often teach data visualization in life sciences, include the visualization of image data. Many students are unsure whether a given image processing step is permissive or even necessary, or if it is misleading audiences. This includes, for example, steps such as changing the default colours in multi-channel microscope images or rotating the field of view. The students' uncertainty also translates into scientific publications; red and green channels are often merged in images, making the resulting figures inaccessible to colourblind audiences. In 2017, I wrote about my observations in a blog post for the Node (https://thenode.biologists.com/color-blind-audiences/photo/) and, following on from this, I initiated an eLIFE ambassador project with Tracey Weissgerber to review image figures. This community project, which focussed on three exemplary and image-heavy research areas (plant sciences, cell biology and physiology), identified common problems of current image figures (Jambor et al., 2021). In addition, it provided a resource of many examples and suggestions for improving image figures.

The concerning numbers of figures that are not fully legible led us to form a working group within the Quality Assessment and Reproducibility for Instruments & Images in Light Microscopy initiative (QUAREP-LiMi) (Boehm et al., 2021; Nelson et al., 2021) dedicated to images in publications. Our working group had two interests. The first was to discuss best practices for reporting image acquisition techniques in the methods sections of published articles. The second aim was to establish a broad consensus for legible microscopy images and image analysis. Initially comprising just four members, the group rapidly grew, engaged in thorough discussions, agreed terminology and formed bridges across many disciplines. Eventually we succeeded in reaching a consensus on what we considered the minimal, recommended and ideal levels of reporting high quality images in scientific figures (Schmied et al., 2023). We delivered the guidelines in easy-to-consume checklists (Fig. 1), which we hope can be used to ensure images fulfil legibility criteria and are fully interpretable in publications. The checklists were primarily designed for light microscopy images, but many points are broad and also apply to other image types, such as photos, electron micrographs and medical images, and to fields outside biology. We envision that the checklists will be used by early career scientists preparing their first publication, as well as by those assessing image quality in publications, such as reviewers, funders or editors. In addition, the guidelines might serve as blueprint for establishing quality criteria for images submitted to repositories, or for publisher's guidelines.

The checklists are the labour of 54 authors and numerous colleagues who provided feedback. The global lockdown of the pandemic was a very fortunate window of opportunity for such a global community initiative: setting up meetings that spanned time zones was eased by the fact that nobody had to commute and that traveling to conferences was on hold. Our consensus guidelines also benefitted from the inclusivity of our community. Everybody, regardless of career stage and occupation, was welcome to join the initiative and contribute towards the initiative, and the meetings were minuted and recorded for those unable to attend. The community was truly global, with members from Melbourne to Madison, and Hangzhou to Heidelberg. The checklists also were enriched by the interdisciplinary dialogue between physicists and biologists, computer scientists and science communicators.

Our image presentation working group within QUAREP-LiMi is by far not the only, and also by no means the first, community effort in the realms of bio-images. On the contrary, the field of bio-imaging seems to have embraced community approaches from the outset. Core microscopy facilities, together with their local and global umbrella organisations, have been key in delivering high quality and shared standards in microscopy training (e.g. Dietzel et al., 2020). These organisations are also in close touch with their user communities, for example to conduct surveys on important topics (Sivagurunathan et al., 2023). Such communities also act as bridges between the branches of bio-imaging, software and hardware microscope specialists. For instance, the European Network of Bioimage Analysts (NEUBIAS; https://eubias.org/NEUBIAS/), a network of experts in life sciences and image data analysis, is relaunching now as a global initiative. They offer training, workshops and concerted publishing of materials (Cimini et al., 2020; Rubens et al., 2020).

Communities not only concern themselves with image acquisition and image data handling. A growing group of scientists are working on ensuring images are available, for example, through image data repositories or archives (Ellenberg et al., 2018; Hartley et al., 2022; Swedlow et al., 2021). Moreover, the Council of Science Editors (CSE) is an international organization of editorial professionals that is dedicated to improving the communication of scientific information, including image communication. Specifically, the CSE has published a white paper with an outline of what constitutes misleading image manipulations (see https://www.councilscienceeditors.org/recommendations-for-promoting-integrity-in-scientific-journal-publications).

The history of science reveals that discussions about image quality are not recent additions to the scientific landscape. Conversations about the value of images have been intertwined with the history of science since their emergence in ancient medical and pharmaceutical atlases (Stückelberger, 1994). Prior to the invention of photography, specimens and microscopic observations were documented through intricate illustrations and plates. These depictions did not merely portray individual specimens but rather encapsulated the collective observations of highly trained experts who possessed the ability to distill fundamental truths about the nature of the specimens (Daston and Galison, 2007). According to Goethe, these experts were tasked with “fixing the empirically variable, excluding the accidental, and eliminating the impure.” In simpler terms, their drawings aimed to represent a specimen type that mirrored the average of all observed variations.

The tradition of documenting specimens through drawings underwent a profound transformation following two developments that revolutionized scientific images. Firstly, the invention of photography and microphotography enabled the technical documentation of specimens. Concurrently, the principle of scientific objectivity, which elevated machines above trained scientists, emerged as a new cornerstone of the natural sciences (Daston and Galison, 2007). An illustrative episode involving physicist Arthur Worthington sheds light on this transformation. A pioneer in fluid dynamics, Worthington meticulously documented the shapes of falling drops through close observation and illustrated his impressions. These illustrations yielded an impressively detailed time-series of the shape of a falling drop, one that was visually pleasing, symmetrical and regular. However, the introduction of photography provided profoundly new insights. Worthington's photographed splashes displayed unexpected irregularities and diverged substantially from the drawings (Fig. 2). Worthington encapsulated his scientific journey in his book, ‘The Romance of Science Series – The Splash of a Drop’ (Worthington, 1895), in which he writes that “the mind of the observer is filled with an ideal splash, whose perfection may never be realized.”

With objective images, new challenges emerged. Early photographs were often coarse-grained and could hardly capture intricate details. In line with the goal of faithfully capturing nature, any form of enhancement of photos was viewed as misleading in a scientific context. However, the boundaries between misleading and necessary improvements were often blurred. Manuals for scientific photography were extensive, encompassing descriptions of filters, lenses, preparation techniques, exposure settings and darkroom manipulations. The manuals included instructions on combining objects from multiple preparations (a practice that is not considered ethical today, but back then was commonly used to save space) but also encouraged detailed captions with the object's source, name, and mode of preparation – information that we now encourage in our checklists. An entire book, ‘La méthode graphique dans les sciences expérimentales’, explained how to make data, including image data, comprehensible through images (Marey, 1885). In sum, Daston dryly writes that “effort and artifice were required to persuade nature to imprint its image” (Daston and Galison, 2007).

Even in the early days of microphotography, the scientific community debated necessary and misleading image manipulations. The renowned scientist Neuhauss expressed concerns about minor alterations, such as changing the background colour, as they often were indicative of more severe image alterations. Yet, while he considered that incising the object or trimming edges could be misleading, he also stated that these changes could sometimes be necessary as “light microscopy depicts everything that does not specifically belong to the object with frightening objectivity...” (Neuhauss, 1890). It is noteworthy to point out that Neuhauss's assertion, “all measures that uniformly impact the entire panel are unequivocally permissible,” remains true today and is included in almost unaltered form in Cromey's authoritative publication on digital images (Cromey, 2013).

Looking back, we can see that debates about image quality are an enduring element of the history of scientific images. Discussions among scientists have accompanied the emergence of scientific atlases, the transition from illustrations to microphotography, and the rise of digital bio-images and their electronic processing. When technologies advance, questions about the authenticity and objectivity of scientific images are asked. In her 2006 article discussing common pitfalls of the emerging field of digital microphotography, Alison North writes that “all data are subject to interpretation” and concludes that only collective efforts ensure that the true picture prevails (North, 2006). Today, we are in the middle of yet another seismic shift with the onset of AI-assisted imaging and image analyses, but at the same time have numerous possibilities for concurrent community debates to accompany these developments.

Over time, the role of scientists has transformed, shifting from the act of distilling a specimen's essence before instructing an illustrator, to becoming technical experts proficient in the art of capturing, communicating and deciphering images. Analogue photographs have transitioned into digital data, and the concept of atlases has resurfaced in the form of image databases (Ellenberg et al., 2018). Photography initially held the promise of aiding scientists in attaining objectivity and reducing human interference in both scientific discovery and replication. Today, the emergence of deep-learning-assisted image processing and analyses once more offers exciting possibilities (Pylvänäinen et al., 2023; Weigert et al., 2018) while also igniting discussions about the boundaries between necessary improvements and misleading alterations in scientific imagery.

The enduring chasm between an observer's perception and an instrument's capabilities of capturing images remains challenging in light microscopy, and also impacts how images and image analyses are presented in scientific publications. A whole chapter of Neuhauss's textbook on microscopy is devoted to reviews of published scientific microscopy images. Alongside excellent microscopic images, Neuhauss also included “inadequate achievements”, which he resurrected “from deserved oblivion” with the intention of sparking community discussion about the good and the bad in microscopy images (Neuhauss, 1890). It is clear – from both historical and recent examples – that we should embrace community efforts and debates as necessary echo-chambers as we collectively strive towards setting standards for insightful scientific images. The foundation, however, will always be the individual researcher striving for truthfulness. Neuhauss writes that “a photograph can only lay claim to objectivity if it is produced by an honest, gifted, micro-photographer, working according to all the rules of the art and richly endowed with patience and skill”.

I would like to thank James P. Sáenz and Christopher Schmied for critically reading this manuscript. I am grateful for a stay at the Institute for Advanced Studies Berlin, which allowed me to explore the history of science of images.